In ITS (Intelligent Tutoring Systems), learner modeling process must be capable to track and change as the learner progresses through the learning process, and his or her knowledge states and behaviors alter. Most of the previous models including BKT and DKT are static or have fixed parameters or a single objective that cannot capture changes in behavior. To address these limitations, this paper presents the Evolutionary Algorithm-Based Adaptive Learner Modeling Framework (EA-ALM), which introduces three key innovations: (1) a dynamic gene encoding mechanism that adjusts knowledge mastery trajectories and behavioral traits in real time, (2) a multi-objective evolutionary optimization process that balances knowledge gain, learning efficiency, and user satisfaction, and (3) a KL divergence-based adaptation trigger that detects behavioral shifts and initiates re-optimization. The evaluation of EA-ALM on STEM dataset (10,500 records from 420 learners) shows that it is more effective than DKT in terms of AUC-ROC with 0.89 (12.3% gains), learning efficiency with 27.6% improvement and the user satisfaction is 4.5 out of 5. Further, it also establishes that EA-ALM it decreases tracking error by 32.1% in contrast to static evolutionary models, which signifies the ability of the algorithm against behavioural variations. However, as highlighted earlier, the issues of scalability and the need to address the learner data datasets for large-scale learning in future using federated learning extensions and distributed implementation. Based on the findings, EA-ALM is a more flexible, human-interpretable, and learner-oriented modeling approach for ITS, which fills the gap between knowledge tracing and behavioral modeling.